Application Of ARMA-SVR Based On Singular Spectrum Analysis In Stock Index Forecasting

Nowadays,the stock market is one of the largest financial markets in the world,predicting the future state of the stock market has always been the focus of the participants in the stock market.However,the stock market is susceptible to a variety of factors,which causes stock prices to fluctuate into a highly unstable time series.In the face of rapidly changing financial market,people have been studying the stock price index and its trend characteristics,therefore,accurate prediction of changes in the stock price index is of great significance for guiding stock market transactions and there are two main aspects:Firstly,from the national perspective,predicting changes in the stock price index can effectively reflect,the volatility and trend of the entire stock market,which contributes to predict the future changes of the national economy and provide a basis for the country to formulate fiscal policies and manage financial investment,thereby effectively avoiding financial risks,strengthening the stability and liquidity of the stock market,and promoting the healthy and sustainable development of the economic undertakings;Secondly,from the perspective of investors,investors can predict the development trend of the stock market based on this,thus to allocate rationally personal assets and select,portfolios with different income levels according to risk preference,so as to obtain high returns while avoiding the hidden risks of the stock market to the greatest extent.As a new nonparametric data-driven technology in time series analysis—singular spectrum analysis(SSA)technology,it gets rid of the limitations of tra-ditional research methods,it constructs the corresponding singular value sequence by creating a time series trajectory matrix and using singular value decomposition in linear algebra to form a corresponding singular value spectrum.Since the time series information reflected by the singular values of different sizes is also differen-t,SSA can decompose the original sequence into the sum of several independent and interpretable components to capture the information of different components of the time series,so it is often used as a pretreatment method for traditional prediction methods.However,in noise reduction,if the noise components are determined artificially by singular value,subjective factors will inevitably be in-volved,which will result in either too much information loss or too much fitting of the sequences after noise reduction,and the prediction is not very accurate.Meanwhile,it is worth noting that the fluctuation caused by the rapid change of noise component is very small,and although the influence on the fluctuation of the whole stock price is weak,the fluctuation can reflect the local change of the stock price in the short term.Therefore,the noise component is not suitable for predicting the long-term trend of stock prices,it can be used to predict short-term changes in stock prices.Based on this,this paper proposes the stock price index prediction mod-el combining autoregressive moving average model(ARMA)and support vector machine regression(SVR)based on singular spectrum analysis,taking full a.dvan-tages of the different prediction model and making more accurate predictions on the short-term trend of stock index sequences.The main steps are as follows:First of all,this paper introduces the principle and modeling process of singular spectrum analysis,support vector machine and autoregressive moving average model;Secondly,on the basis of the above model,the original data sequence is decomposed into three parts:trend sequence,fluctuation sequence and noise se-quence by applicating the singular spectrum analysis technology.Then perform stationarity test on the obtained subsequences,the SVR model is used to predic-tnon-stationary sequences and the ARMA model is used to predict stationary sequences,and final prediction result is obtained by integrating the above result-s;Finally,this paper compares with the three support vector regression models based on singular spectrum analysis,and compares the prediction effect according to different model evaluation criteria.The results show that there is still useful information in the noise sequence extracted by the singular spectrum analysis method.Therefore,it is necessary to preserve the noise sequence and analyze it.The subsequence structure obtained by singular spectrum analysis becomes simpler and easier to be simulated.The proposed combination model of ARMA and SVR based on singular spectrum analysis is more accurate than the single model of SVR based on singular spectrum analysis.